MISSION-ORIENTED POLICIES AND THE "ENTREPRENEURIAL STATE" AT WORK: AN AGENT-BASED EXPLORATION - Giovanni Dosi Francesco Lamperti Mariana Mazzucato ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
MISSION-ORIENTED POLICIES AND THE “ENTREPRENEURIAL STATE” AT WORK: AN AGENT-BASED EXPLORATION Giovanni Dosi Francesco Lamperti Mariana Mazzucato Mauro Napoletano Andrea Roventini SCIENCES PO OFCE WORKING PAPER n° 14/2021
EDITORIAL BOARD
Chair: Xavier Ragot (Sciences Po, OFCE)
Members: Jérôme Creel (Sciences Po, OFCE), Eric Heyer (Sciences Po, OFCE), Sarah Guillou (Sciences
Po, OFCE ), Xavier Timbeau (Sciences Po, OFCE)
CONTACT US
OFCE
10 place de Catalogne | 75014 Paris | France
Tél. +33 1 44 18 54 24
www.ofce.fr
WORKING P AP ER CITATION
This Working Paper:
Giovanni Dosi, Francesco Lamperti, Mariana Mazzucato, Mauro Napoletano and Andrea Roventini
Mission-Oriented Policies and the “Entrepreneurial State” at Work: An Agent-Based Exploration
Sciences Po OFCE Working Paper, n° 14/2021.
Downloaded from URL: www.ofce.sciences-po.fr/pdf/dtravail/WP2021-14.pdf
DOI - ISSN
© 2021 OFCEABOUT THE AUTHORS
Giovanni Dosi, Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna
Email Address: g.dosi@santannapisa.it
Francesco Lamperti, Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna and RFF-CMCC European
Institute on Economics and the Environment
Email Address: (f.lamperti@santannapisa.it
Mariana Mazzucato, Institute for Public Purpose and Policy, University College London
Email Address: m.mazzucato@ucl.ac.uk
Mauro Napoletano, GREDEG, CNRS, Université Côte d’Azur, SKEMA Business School, Sciences Po, OFCE and Scuola
Superiore Sant’Anna
Email Address: mauro.napoletano@univ-cotedazur.fr
Andrea Roventini, Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna and Sciences Po, OFCE
Email Address: a.roventini@santannapisa.it
ABS TRACT
W e study the impact of alternative innovation policies on the short- and long-run performance of the economy, as well as
on public finances, extending the Schumpeter meeting Keynes agent- based model (Dosi et al., 2010). In particular, we
consider market-based innovation policies such as R&D subsidies to firms, tax discount on investment, and direct policies
akin to the “Entrepreneurial State” (Mazzucato, 2013), involving the creation of public research-oriented firms diffusing
technologies along specific trajectories, and funding a Public Research Lab conducting basic research to achieve radical
innovations that enlarge the technological opportunities of the economy. Simu- lation results show that all policies improve
productivity and GDP growth, but the best outcomes are achieved by active discretionary State policies, which are also able
to crowd-in private investment and have positive hysteresis effects on growth dynamics. For the same size of public
resources allocated to market-based interventions, “Mission” innovation policies deliver significantly better aggregate
performance if the government is patient enough and willing to bear the intrinsic risks related to innovative activities.
Keywords
Innovation policy, mission-oriented R&D, entrepreneurial state, agent-based modelling.
J EL
O33, O38, O31, O40, C63.Mission-Oriented Policies
and the “Entrepreneurial State” at Work:
An Agent-Based Exploration∗
Giovanni Dosi Francesco Lamperti Mariana Mazzucato
Mauro Napoletano Andrea Roventini
May 21, 2021
Abstract
We study the impact of alternative innovation policies on the short- and long-run performance
of the economy, as well as on public finances, extending the Schumpeter meeting Keynes agent-
based model (Dosi et al., 2010). In particular, we consider market-based innovation policies such as
R&D subsidies to firms, tax discount on investment, and direct policies akin to the “Entrepreneurial
State” (Mazzucato, 2013), involving the creation of public research-oriented firms diffusing tech-
nologies along specific trajectories, and funding a Public Research Lab conducting basic research
to achieve radical innovations that enlarge the technological opportunities of the economy. Simu-
lation results show that all policies improve productivity and GDP growth, but the best outcomes
are achieved by active discretionary State policies, which are also able to crowd-in private invest-
ment and have positive hysteresis effects on growth dynamics. For the same size of public resources
allocated to market-based interventions, “Mission” innovation policies deliver significantly better
aggregate performance if the government is patient enough and willing to bear the intrinsic risks
related to innovative activities.
Keywords: innovation policy, mission-oriented R&D, entrepreneurial state, agent-based modelling.
JEL codes: O33, O38, O31, O40, C63.
∗
Giovanni Dosi (g.dosi@santannapisa.it), Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna, (Pisa,
Italy); Francesco Lamperti (f.lamperti@santannapisa.it), Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna,
(Pisa, Italy) and RFF-CMCC European Institute on Economics and the Environment (Milan, Italy); Mariana Maz-
zucato (m.mazzucato@ucl.ac.uk), Institute for Public Purpose and Policy, University College London (London, UK);
Mauro Napoletano (mauro.napoletano@univ-cotedazur.fr), GREDEG, CNRS, Université Côte d’Azur, and SKEMA Business
School (Nice, France), and Sciences Po, OFCE (France) and Scuola Superiore Sant’Anna, (Pisa, Italy); Andrea Roventini
(a.roventini@santannapisa.it), Institute of Economics and EMbeDS, Scuola Superiore Sant’Anna, (Pisa, Italy) and Sciences
Po, OFCE (France). Corresponding author: Francesco Lamperti, email: f.lamperti@santannapisa.it; postal address: Institute
of Economics, Scuola Superiore Sant’Anna, piazza Martiri della Libertà 33, 56127 Pisa, Italy.
11 Introduction
In this paper, we extend the Schumpeter meeting Keynes agent-based model (Dosi et al., 2010) to assess
the impact of different innovation policies on the short- and long-run performance of the economy, as
well as on the public budget.
The stagnating aftermaths of the Great Recession and, more recently, of the COVID-19 pandemics,
call for public policies able to restore robust economic growth. Such crises also exacerbated the pre-
existing productivity slowdown experienced by most developed economies. This implies that govern-
ment should introduce policies to influence the pace of innovation and technological change, which are
the major drivers of long-run economic growth. The Next Generation EU program released by the Eu-
ropean Commission goes explicitly in this direction. However, in our view, the contemporary discourse
on innovation policies has been far too narrow, quite disjoint from their implications for the economic
and social future of our societies. In fact, it is remarkable that, in the past, some of the most important
“innovation policies” were not called as such. The Manhattan Project, the Apollo Program, Nixon’s
“war on cancer” were not discussed, if at all, as “policies” but as major societal objectives, well shielded
from the narrow concerns of economists’ cost-benefit analyses. On the contrary, nowadays, innova-
tion policies – except for war-related innovations and pandemic emergencies - have to pass through the
dire straits of efficiency criteria. However, even on these narrower grounds, we shall show, innovation
policies are well worth.
Innovation policies (written large, and meant to include science and technology policies) broadly
refer to the design of a variety of instruments aimed at generating new knowledge, new products and
more efficient production techniques (within an enormous literature, see from Bush et al., 1945 to Free-
man and Soete, 1997; Edler and Fagerberg, 2017; Criscuolo et al., 2020). Depending on the type and
scope of the policy tools employed, innovation policy might require more or less extensive involve-
ment of the public sector in the economy. A broad distinction is between indirect and direct innovation
policies (Dosi, 1988; Dosi and Nelson, 2010; Mazzucato and Semieniuk, 2017). Indirect policies tend to
be “market-friendly” as they provide monetary incentives to firms to improve their innovative perfor-
mance (e.g. R&D subsidies) or to speed-up their technological renewal (e.g., investment tax discounts).
In an influential debate at the OECD in the early 80s, they were called “diffusion-oriented” policies (Er-
gas, 1987). Differently, direct innovation policies imply an active of role of the public sector in shaping
the rates and directions of innovative activities, which means - to paraphrase Nelson (1962) - shap-
ing technological landscape and search regimes, taking risks that private businesses are not willing to
sustain, and pursuing pathbreaking technological developments. Direct innovation policies respond to
Freeman (1987) plea for policies creating systems and institutions able to nurture the generation and
diffusion of new knowledge across the economy, the creation of new industries and markets and - ul-
timately - to fuel economic growth. These policies may certainly be facilitated by an Entrepreneurial
State (Mazzucato, 2013) that takes the lead and directly invests in the search for novel technological
opportunities (possibly directed to specific missions; see also Mazzucato, 2018a and Mazzucato, 2021).
The ability of alternative innovation policies to spur innovation, crowd in private investment and
2deliver sustained long-run growth is highly debated. Notwithstanding a large body of studies evaluat-
ing single policies (see Becker, 2015, for a survey), systematic comparisons of policy designs are scarce
in the literature (Grilli et al., 2018), especially from a macroeconomic perspective (Di Comite and Kancs,
2015). A recent review by Bloom et al. (2019) discusses pros and cons of various instruments, suggesting
a trade-off between the short run, where tax incentives and subsides are effective in stimulating inno-
vation, and long run outcomes, which would benefit from systemic investments in universities and
education. However, Bloom and co-authors overlook (or dismiss) direct policies, based on the argument
that the effects of these policies are hard to be identified econometrically. In addition, those policies,
it is suggested, lack an economic rationale - of course in terms of the conventional economic theory,
according to which were it not for market failures and externalities, one better leave the market and
the search for innovations to itself.
In this work, we shall indeed show the robust rational of direct policies in complex evolving economies.
We extend the Schumpeter meeting Keynes (K+S) macroeconomic agent-based model (Dosi et al., 2010)
to systematically compare the impact of direct and indirect innovation policies on economic perfor-
mance, while accounting for their impact on the public budget. 1 In that, the paper also contributes to
the literature about modelling of R&D, innovation activities and their impacts on the macroeconomy,
integrating the representation of technological change, its sources and consequences within an agent-
based perspective (for germane contributions see Russo et al., 2007; Dawid et al., 2008; Lorentz et al.,
2016; Caiani et al., 2019; Dosi et al., 2019; Fagiolo et al., 2020 and the survey in Dawid, 2006).
The K+S model is composed of two vertically-related sectors, wherein heterogeneous firms strive to
develop new technologies and locally interact by exchanging capital-goods in a market with imperfect
information. This is the Schumpeterian engine of the model: new machine tools are discovered and
diffuse within the economy both via imitation activities of competing capital-good producers and via
investment by consumption-good firms. Firm investment depends on firm demand expectations, as
well as on their financial conditions and it constitutes, together with worker consumption and public
expenditures, the Keynesian soul of the model. Aggregate demand dynamics in the model affects not
only business cycles, but also the pace of technological change (see e.g. Dosi et al., 2016). The K+S
model is therefore able to go beyond the traditional separation between “coordination" and “change" in
economics (Dosi and Virgillito, 2017).
Indeed, the K+S family of models represents flexible environments which can be used as virtual lab-
oratories for policy experiments to investigate a variety of policy interventions and perform counter-
factual analyses. We examine four innovation policy regimes and their possible combinations, namely
(i) R&D subsidies to capital-good firms; (ii) tax-discounts on consumption-good firms’ investments; (iii)
the creation of a public research-oriented capital-good firm; (iv) the institution of a National Research
Laboratory which tries to discover radical innovations that enlarge the set of technological opportu-
nities available in the economy. The first two experiments mimic indirect innovation policies, while
1
Agent-based models are particularly suited to evaluate different combinations of policies in frameworks characterized by
deep uncertainties, technical and structural change. More on that in Fagiolo and Roventini (2017); Dosi and Roventini (2019);
Dawid and Delli Gatti (2018). We also suggest to look at Dosi et al. (2020) for a systematic comparison of market-based and
industrial policies in fostering catching-up.
3the latter pair captures key features of direct or “Entrepreneurial-State” policies. Finally, we consider
a benchmark scenario where the public resources is used to support private consumption instead of
innovation policies.
Simulation results show remarkable differences across innovation policy regimes. First, all inno-
vation policies spur productivity and GDP growth, but to different degrees, while this is not the case
for transfers to households. Second, the impact of direct innovation policies is larger vis-à-vis indirect
ones and entails effects of positive hysteresis (Dosi et al., 2018; Cerra et al., 2021) putting GDP on higher
growth trajectories. However, Entrepreneurial-State policies are risky: their positive impact tend to
show up on longer time horizons as compared with indirect interventions, and they can fail to dis-
cover new technologies. Nonetheless, extensive Monte Carlo analyses show that, on average, direct
innovation policies deliver higher productivity and GDP growth, while being less expensive in terms
of net public resources, compared to “indirect” forms of intervention. The impact of Entrepreneurial-
State interventions is stronger when they combine the presence a public firm with a National Research
Laboratory. Conversely, indirect monetary incentives tend to be associated with some redundancy –
that is transfer of resources to firms with little effect on the intensity of search. Finally, all innovation
policies we consider crowd in private R&D investment (in line with Moretti et al., 2019 and Pallante
et al., 2020), although direct interventions provide, again, the most bang for their buck. Accordingly,
our results suggest that the type of tools utilised by a mission-oriented Entrepreneurial State (Mazzu-
cato, 2013, 2018a, 2021) are also more effective at meeting uncontroversial innovation policy goals of
productivity and growth gains.
To sum up, our results indicate that innovation policies are highly effective. In particular, when pub-
lic resources are concentrated on clear missions and Entrepreneurial-State interventions, they appear
to deliver large gains in economic performance compared to policies based on monetary incentives.
This should be taken into account by policy makers when designing vast policy plans such as the Next
Generation EU to jump-start growth in economies hardly hit by the COVID-19 crisis.
The rest of the paper is organized as follows. Section 2 provides a critical overview of the literature
on innovation policies. In Section 3, the K+S model is introduced. The empirical validation of the model
is performed in Section 4. In Section 5, we present the results of innovation policy experiments. Finally,
Section 6 concludes the paper.
2 Innovation policies: a critical review
Economic theory identifies innovation as the most relevant driver of industrial development, special-
ization and long-run economic growth. This holds true both in neoclassical (Solow, 1957; Romer, 1986;
Aghion and Howitt, 1992) and evolutionary theories (Nelson and Winter, 1982; Dosi, 1982; Dosi et al.,
1994). However, the underlying views about how knowledge evolves, accumulates, diffuses and - ulti-
mately - affects productivity are profoundly different across these two theoretical paradigms (see Dosi,
1988; Dosi and Nelson, 2010, among others). Such differences also often map into opposing prescrip-
tions with respect to innovation policy.
4We define innovation policies, to repeat, as the set of attempts carried out by a government to
shape or influence the generation and diffusion of new knowledge and new technologies. All this can
be implemented either via monetary instruments, regulations or direct interventions, often but not
always with the purpose of increasing productivity and economic growth. Some other times, they can
be just be an unintended consequence of policies meant to achieve other purposes – e.g. winning a war
(Moretti et al., 2019; Gross and Sampat, 2020). But what motivates innovation policies themselves?
The market view closely based on a neoclassical perspective basically justifies policies only in pres-
ence of market failures or untraded externalities. Assuming in a first approximation the equivalence
between technological knowledge and information, the latter has an intrinsic public-good nature, im-
plying an endemic tendency to underinvest in expensive activities of search by private profit-motivated
agents (Arrow, 1951, 1962), which can be mitigated by various forms of transfers and incentives. An-
other way to partially align incentives to innovate by private actors and social objectives – as a good
deal of the current narrative goes, vastly overstretching the implication of Arrow’s argument – entails
the deepening and strengthening of Intellectual Property Rights (IPR) thus supposedly increasing the
equilibrium rates of allocation to R&D investments, etc. There are many reasons why this argument is
very weak.
Let us start with the “market failure” approach, appealing as it is for its simplicity, which indeed con-
tinues to be influential among policy makers (OECD, 2010; EU, 2020) and economists alike. However, it
is theoretically flawed and empirically unfounded. On the theoretical side, the argument is postulated
on the difference between ideal “complete markets” and actual ones. But if this is so, the whole world
is a huge market failure: there is hardly any market which looks like a complete market in the Arrow-
Debreu sense (more in Stiglitz, 1996; Cimoli et al., 2009)! Indeed, there is a fundamental incompatibility
between innovation and general equilibrium, basically for two reasons. First, if an innovation is a true
innovation, one cannot know about it ex ante, otherwise it would not be an innovation: therefore it
is also impossible to attribute probabilities to its occurrence, let alone having “rational expectations”
about them and their mapping into expected costs. Thus, markets must be incomplete by definition.
Second, the very presence of technological knowledge (Arrow calls it “technical information”) implies
an extreme form of increasing returns and thus ubiquitous non-convexities, multiple equilibria, or non-
existence of equilibrium at all (see Arrow, 1996 and the comments by Arrow in Teece, 2019). Of course,
with that disappears all the welfare properties of general equilibrium, taken for granted in “market
failure” evaluations. Analytically, “computable general equilibria”, often used to plug innovation into
aggregate models, in this context, is just an oxymoron. The “market failure argument” is mislead-
ing also for other reasons. Indeed, the equivalence between knowledge and information is just a first
rough approximation: while information can be easy to access, the same does not necessarily hold
for knowledge. Not all knowledge can be codified: much economically useful knowledge is tacit and
heterogeneously distributed across actors and contexts (Polanyi, 1944; Nelson and Winter, 1982; Dosi,
1988; Winter, 1998; Metcalfe, 2005; Dosi and Nelson, 2010).
More generally, the empirical evidence supporting any link between incentives and propensity to
innovate is at best fuzzy. First, the empirical evidence supporting the effectiveness of monetary subsi-
5dies and stronger IPR regimes to stimulate private R&D spending is rather weak (Zúñiga-Vicente et al.,
2014; Dimos and Pugh, 2016; Papageorgiadis and Sharma, 2016), despite the fact that these policies typ-
ically entail large fiscal costs. Indeed, firms might tend to keep their R&D steady and simply exploit
public subsidies and tax-credits to boost their profits (Marino et al., 2016; Mohnen et al., 2017). Second,
stronger IPRs might not matter significantly in firm-level decisions and can even decrease the long-run
pace of innovation (Dosi et al., 2006; Dosi and Stiglitz, 2014; Cimoli et al., 2014; Stiglitz, 2014). For ex-
ample Levin et al. (1987), Fagerberg (2017) and Cohen (2010) show that in most industries firms are not
much concerned about the lack of strong IPR as the capabilities underpinning their innovative perfor-
mance cannot be copied easily (Dosi and Nelson, 2010; Edler and Fagerberg, 2017). On the contrary,
many firms have close interactions and knowledge exchanges with relevant parties (e.g., customers,
suppliers, universities, public research institutions, etc.) which nurture the transfer of tacit knowledge
during the innovation process.
Furthermore, the market failure approach is even less useful when radical technological change is
needed (see Mazzucato, 2016). Private businesses tend to invest in new technologies only after the high
risks and uncertainty have been absorbed by research and development activities directly funded by
the public sector. In this case, mission-oriented policies are needed to create new technologies, new
sectors and new markets (Foray et al., 2012). Such innovation policies consider the public sector as
an Entrepreneurial State mostly engaged in industry creation and market shaping rather than market
fixing, actively setting new innovation directions towards significant social goals (missions). The idea
of market shaping and mission-orientation has began to gained acceptance in recent years in Europe
where it seems to be adopted by the European Commission - in relation to grand societal challenges
such as the green transition (Mazzucato, 2018b, 2019). This finally reflects disappointment in the ability
of market fixing approaches to address these challenges and recognition that the appetite for risk, long
term thinking and capacity for coordination in the private sector is inadequate for producing a deci-
sive shift in the direction of innovation (Mazzucato and Semieniuk, 2017, 2018). Public policies must
therefore support all phases of the innovation process, taking risks (and possible losses) that the private
sector will not absorb, waiting patiently for the rewards of innovation and coordinating activities across
public and private stakeholders (Mazzucato, 2013). Perhaps less widely acknowledged is the economic
case for a mission-oriented Entrepreneurial State. The economic impacts from such policies are often
hard to quantify empirically, being associated with dynamic spillovers, even when the social ones are
quite obvious. A priori, we would expect Entrepreneurial State policies to have high potential for gen-
erating growth due to the fact they target new markets, technologies and directions of discovery. This
means they have also the potential to create opportunities for advancement in productivity, consumer
demand, international competitiveness and so forth which would not be created by the private sector
alone (Mazzucato, 2013, 2018a).
The historical record provides compelling cases in support of this. Governments invested directly
in the technologies that enabled the emergence of mass production and IT revolutions and undertook
the bold policies required to deploy them throughout the economy (Block and Keller, 2015; Ruttan,
2006). Many of the examples of this relate to the pervasive impact of military and space innovation (the
6Manhattan project, the Apollo program and ARPANET - the progenitor of the internet - are among
the most famous; see Gross and Sampat, 2020, 2021) but, more recently, successful results have been
highlighted across many other technological landscapes, including the biotechnology industry (Lazon-
ick and Tulum, 2011), nanotechnologies (Motoyama et al., 2011), and the emerging clean-tech sector
(Mazzucato, 2015; Steffen et al., 2020).
Beyond the selection of the missions to pursue, which reflects broader societal and political objec-
tives, the Entrepreneurial State approach to innovation policy can be summarized across three defining
features (Mazzucato, 2016). First, public organizations should experiment, conduct research, learn and
take risks. Second, policy design should create symbiotic private–public partnerships, overtaking the
idea of de-risking the private investment and fostering a collaborative environment, characterized by
joint R&D projects to create new products and services (e.g., new vaccines; Chataway et al., 2007),
and crowding-in of private investment (see for example Engel et al., 2016; Moretti et al., 2019; Pallante
et al., 2020). Finally, it should provide a system of rewards for the public sector to ensure the long run
sustainability of the high risk-taking investments described above, as well as for public accountability
purpose.
One of the implications of the “market failure approach” is that it calls for the state to intervene
as little as possible in the economy, in ways that minimize the risk of “government failures”, whatever
that means in complex evolving economies. A corollary is also the drive to outsource the innovation
process from public organizations to private firms.
To the contrary, a mission-oriented Entrepreneurial State aims at shaping the direction of tech-
nological change, employing a mix of indirect instruments (schemes of incentives) and, much more
important, direct interventions (e.g. through public agencies, formal public-private collaborations, use
of public banks to finance bold R&D projects), and coordinating the governance of the whole inno-
vation chain.2 Under this perspective, the State should not limit itself to provide funding for basic
knowledge and help protecting innovation through implementation of IPRs, as the market failure the-
ory would suggest, but also identify and rectify such systemic problems coordinating all levels of public
administration and private stakeholders (Metcalfe, 1994, 1995; Edquist, 2011).
3 The K+S model
We investigate which type of innovation policies is more effective in stimulating innovation, produc-
tivity and output growth in the Schumpeter meeting Keynes model extended to account for radical
innovations and the variable cost of public debt (Dosi et al., 2010, 2013).3 Our stylized representation
2
In this respect, various similarities are shared with the so-called “system-oriented” innovation policies (Edler and Fager-
berg, 2017), which builds on the literature on National (Freeman, 1987; Lundvall, 1988, 2010) and Sectoral (Malerba, 2002)
Innovation Systems and looks at the systemic nature of the innovation process as emerging from the interaction of a number
of factors, including knowledge, skills, financial resources, demand etc. When the system does not sufficiently provide for
those factors or fails at coordinating them, a “system failure” may hamper innovation activity.
3
See also Dosi et al. (2017) for a survey about the Schumpeter meeting Keynes family of models. Indeed, the K+S model has
been extended to account for multiple banks and fiscal-monetary policy trade-offs (Dosi et al., 2015), decentralized interactions
in the labour market (Dosi et al., 2017, 2021) and the coupled dynamics of climate climate and the economic growth (Lamperti
7of an economy is composed of a machine-producing sector composed of F1 firms, a consumption-good
sector composed of F2 firms, an ecology of consumers/workers, and a public sector. Capital-good firms
invest in R&D and produce heterogeneous machines. Consumption-good firms combine machine tools
bought by capital-good firms and labour in order to produce a final product for consumers. The public
sector levies taxes on firms’ profits, pay unemployment benefits, and implement the selected innovation
policies.
3.1 Innovation and technological progress
The Schumpeterian engine of the K+S model stems from the innovation and imitation search of capital-
good firms, which produce machine-tools using labour only. The technology of the machines of vintage
τ is captured by the couple of coefficients (Ai,τ , Bi,τ ), where the former represents the productivity
of machines employed in the consumption-good industry, while the latter indicates the productivity of
the production technique needed to manufacture the machine. Given the monetary wage, w(t), paid
to workers, the unitary cost of production of capital-good firms is given by:
w(t)
ccap
i (t) = . (1)
Bi,τ
Similarly, the “quality” of the machines captured by (Ai,τ ) defines the unitary production cost of
consumption-good firms (indexed by j):
w(t)
ccon
j (t) = . (2)
Ai,τ
Capital good firms adaptively strive to increase market shares and profits trying to improve their
technology via innovation and imitation. They are both costly processes: firms invest in R&D a fraction
of their past sales in the attempt to implement incrementally new technologies, discover radically new
innovations and imitate more advanced competitors. More specifically,
RDi (t) = υSi (t − 1), υ ∈ {0, 1} (3)
indicates firm i’s spending in R&D, which is split into in-house (incremental) innovation (INi ) and
imitation (IMi ) activities:
INi (t) = ξRDi (t), IMi (t) = (1 − ξ)RDi (t), ξ ∈ [0, 1]. (4)
As in Dosi et al. (2010), innovation and imitation are depicted as two-steps processes. The first step
captures firms’ search for new technologies through a draw from a Bernoulli distribution, wherein the
real amount invested in R&D (i.e. the number of hired researchers) positively affects the likelihood of
success. More precisely, the parameters controlling the likelihood of success in the Bernoulli trial for
et al., 2018a, 2019, 2020, 2021).
8the innovation and imitation process, θIN (t) and θIM (t) respectively, correspond to:
θIN (t) = 1 − e−oIN INi (t) , oIN > 0, (5)
θIM (t) = 1 − e−oIM IMi (t) , oIM > 0; (6)
where the parameters 0 < −oIN , oIM 6 1 capture the the search capabilities fo firms.
The second step differs for innovation and imitation activities. Let us consider innovation first.
Successfully innovating firms will access a new technology, whose technical coefficients are equal to:
Ai,τ +1 = Ai,τ (1 + χA,i ) (7)
Bi,τ +1 = Bi,τ (1 + χB,i ) (8)
where χA,i and χB,i are independent draws from a Beta(α, β) distribution over the support [ξ1 , ξ2 ],
with ξ1 < 0 and ξ2 > 0. The support captures the technological opportunities available for the firms.
Note that as χ(t) is allowed to be negative, the newly discovered technology may be inferior to the
current one. This reflects the intrinsic trial and error process associated to any search for new tech-
nologies.
Successful imitators have the opportunity to copy the technology (embodied in the two technical
coefficients A and B) of one of their competitors. The imitation probability negatively depends on the
technological distance between each pair of firms. More precisely, the technological space is modelled
as a 2-dimensional Euclidean space (A, B), where `2 is chosen as the metric determining distance
between couples of points:
q
T Di,j = (Ai − Aj )2 + (Bi − Bj )2 , (9)
where the vintage of the technology employed by firms i and j is dropped to ease notation. For each
imitator, competitors are ranked according to their (normalized) technological distance N T Di,j =
T Di,j / j T Di,j and a draw from a uniform distribution on the unitary interval determines the firm
P
whose technology will be imitated.
When a novel technology is developed or imitated, firms decide whether to adopt it or not by
comparing its overall costs through the following routine:
min[phi (t) + bccon,h ] h ∈ {in, im, τ }, (10)
where b is a payback parameter (more on that in Section 3.2), p is the price of the machine and c is
the unitary production cost a firm would incur in employing the imitated (im), newly developed (in)
or incumbent (τ ) technology. Once the machine to put in production is selected, capital-firms fix the
price as a contant mark-up on their unit cost of production. The capital-good market is characterized
by imperfect competition: capital-good firms advertise their product to their historical customers, as
well as to a subset of potential new ones.
9Figure 1: Shift of technological opportunities implied by radical innovations
previous
distribution
novel
distribution
Technological opportunities
Beyond in-house incremental innovations and imitation, we allow for the discovery of radical inno-
vations, which are intended here as innovations that change the technological landscape and increase
the technological opportunities available in the economy. Examples of such radical innovations include
electricity, energy storage and the Internet. Following the lines of Mazzucato (2013), these innovations
are rarely the outcome of a single research project within private businesses, but more likely depend on
a broader, systemic effort encompassing both public (from basic to applied) and private research, often
carried out through private-public collaborations and characterized by sequences of trials and errors
(see also Mowery, 2010; Block and Keller, 2015 and the discussion in section 2). To capture these fea-
tures, we model radical innovations as shifts of the support [ξ1 , ξ2 ] of the distribution of technological
opportunities available to the firm (see also Figure 1):
ξ1RI = ξ1 + χRI , ξ2RI = ξ2 + χRI . (11)
The probability of discovering a radical innovation depends positively on the cumulative R&D ex-
penditures performed by the capital-good firm (CRDi ) and by public research agencies (CRDpublic ).
Private cumulative R&D, CRDi (t) = s 0 and η2 > 0 controlling the shape of the logistic function. Indeed, there is robust evidence
supporting a non-linear positive association between a sufficiently large stock of cumulated knowl-
edge and the discovery of breakthrough innovations (Phene et al., 2006; Dunlap-Hinkler et al., 2010;
Kaplan and Vakili, 2015). Enlarged technological opportunities diffuse though the capital-good sector
via the imitation of competing firms. However, radical innovations are more difficult to copy as they
increase the technological distance between the firm mastering the new state-of-the-art technology
and its competitors.
103.2 Investment and technological diffusion
Firms in the consumption-good industry produce a homogeneous good using their stock of machines
and labor under constant returns to scale. They invest to expand their capital stock and/or to replace
their obsolete machines with new ones. Note that such investments contribute to the technological
diffusion of state-of-the-art technologies in the economy.
Let us first consider expansionary investment. Firms face a demand created by the expenditures of
workers, and plan their production according to (adaptive) expectations over such a demand, desired
inventories, and their stock of inventories.4 Whenever the capital stock is not sufficient to produce the
desired amount, firms invest (EIj ) in order to expand their production capacity:
EIj (t) = Kjd (t) − Kj (t), (13)
where Kjd and K denote the desired and actual capital stock respectively.
Further, firms invest to replace current machines with more technologically advanced ones accord-
ing to a payback period routine. In a nutshell, they compare the benefits entailed by new vintages
embodying state-of-the-art technology vis-á-vis the cost of new machines, taking into account the
horizon in which they want to recover their investment. In particular, given the set of all vintages of
machines owned by firm j at time t, the machine of vintage τ is replaced with a new one according to:
pnew pnew
= ≤b (14)
ccon new
h i
j (t) − c
w(t)
− cnew
Ai,τ j
where pnew and cnew are the price and unitary cost of production associated to the new machine and b
is a parameter capturing firms’ “patience” in obtaining net returns on their investments.5 The vintages
of machines that satisfies Eq. 14 constitute the replacement investment of the firm, SIj (t). Aggregate
investment just sums over the investments of all consumption good firms:
Ij (t) = EIj (t) + SIj (t). (15)
As the capital-good market is characterized by imperfect information, consumption-good firms
choose their capital-good supplier comparing price and productivity of the currently manufactured
machine-tools. The model thus entails local interaction among heterogeneous suppliers and customers.6
Consumption-good firms sets the price of their final good applying a variable mark-up on their
4
In the benchmark setup, expectations are myopic. The results are robust for different expectation formation mechanisms.
More on that in Dosi et al. (2020).
5
Our assumptions are in line with a large body of empirical literature showing that replacement investment is typically
not proportional to the capital stock, but a crucial strategic decision of firms (see e.g. Feldstein and Foot, 1971; Eisner, 1972;
Goolsbee, 1998).
6
More on that in Dosi et al. (2010). Note also that machine production is a time-consuming process: consumption-good
firms receive the ordered machines at the end of the period. This is in line with a large body of literature: see, e.g., Rotemberg
(2008) for details on pricing, imperfect information and behavioural attitudes of consumers and Boca et al. (2008) for the
presence of gestation lag effects in firms’ investments.
11unit cost of production. In line with the evolutionary literature and a variety of “customer market”
models (Phelps and Winter, 1970), the mark-up changes over time according to the evolution of firm’s
market shares: firms increase prices if their market share is rising and decrease it when the market
share falls. Consumers have imperfect information regarding the final product (see Rotemberg, 2008
for a survey on consumers’ imperfect price knowledge) which prevents them from instantaneously
switching to the most competitive producers. For this reason, market competition is captured via a
replicator dynamics: the market share of firms more competitive than the industry average increases,
while that of less competitive ones shrinks over time. Firms’ competitiveness depends on their price
and on their capacity to satisfy demand in the past.7
At the end of each period, consumption-good and capital-good firms compute their profits and
update their stock of liquid assets. Firms with zero market shares or negative net assets die and a new
breed of firms enters the market. Overall, the number of firms is fixed, hence any dead firm is replaced
by a new one. In line with the empirical literature on firm entry (Bartelsman et al., 2005), we assume that
entrants are on average smaller than incumbents, with the stock of capital of new consumption-good
firms and the stock of liquid assets of entrants in both sectors being a fraction of the average stocks of
the incumbents. Concerning the technology of entrants, new consumption-good firms select amongst
the newest vintages of machines, while the technology of new capital-good firms is on average worse
than incumbents’ ones.
3.3 The public sector and the macroeconomic framework
Workers-consumers have a marginal propensity equal to one in the model. Accordingly, aggregate
consumption (C) is computed by summing up over the income of both employed and unemployed
workers:
C(t) = w(t)LD (t) + wU [LD (t) − LS (t)], (16)
where w represent wages, wU the unemployment subsidy and LD and LS labour demand and labour
supply respectively. Wages are linked to the dynamics of productivity, prices and unemployment rate
by the following wage equation:
¯
∆AB(t) ∆cpi(t) ∆U (t)
w(t) = w(t − 1) 1 + ψ1 ¯ + ψ2 + ψ3 , (17)
AB(t − 1) cpi(t − 1) U (t − 1)
where AB
¯ indicates the average productivity in the economy, cpi is the consumer price index and
U stands for unemployment rate. The labor market does not necessarily clear and both involuntary
unemployment and labor rationing can occur.
The unemployment subsidies - a fraction of the current market wage - are paid by the public sector
(G indicates such spending), which also levies taxes on firm profits. Taxes and subsidies are the fiscal
leverages that contribute to the aggregate demand management regimes. Further, the government can
7
Unfilled demand is due to the difference between expected and actual demand. Firms set their production according to
the expected demand. If a firms is not able to satisfy the actual demand, its competitiveness is accordingly reduced. On the
contrary, if expected demand is higher than actual one, inventories accumulate.
12run innovation policy incurring in additional spending as indicated by IP (more on that in section 5).
The deficit is then equal to:
Def (t) = G(t) − T axes(t) + CD(t) + IP (t), (18)
where CD indicates the cost of public debt (i.e. interests on previous debt) and satisfies CD(t) =
rpd (t)P D(t − 1), with P D denoting the stock of public debt and rpd (t) the interest rate. Differently
from Dosi et al. (2010), the interest rate on government bonds changes over time according to the ratio
between the public debt and GDP:
P D(t)
rpd (t + 1) = rpd (t) + % , (19)
GDP (t)
with % > 0. The above assumption allows one to capture the long-run cost of innovation policies and
the possible emergence of vicious debt cycles triggered by excessive public expenditures.8
Finally, the model satisfies the standard national account identities: the sum of value added of
capital- and consumption goods firms equals aggregate production that, in turn coincides with the sum
of aggregate consumption, investment and change in inventories:
X X
Qi (t) + Qj (t) ≡ Y (t) ≡ C(t) + I(t) + ∆N (t), (20)
i j
where Qi and Qj represent the production of capital and consumption good firms respectively and ∆N
stands for the variation of inventories.
4 Simulation set-up and empirical validation
The foregoing model does not allow for analytical, closed-form solutions. This is a distinctive feature
of many ABMs that stems from the non-linearities present in agent decision rules and their interac-
tion patterns, and it implies running computer simulations to analyze the properties of the stochastic
processes governing the coevolution of micro and macro variables (more on that in Windrum et al.,
2007; Fagiolo and Roventini, 2017; Fagiolo et al., 2019). In what follows, we therefore perform exten-
sive Monte-Carlo analyses to wash away cross-simulation variability. More precisely, all results are
presented either as single simulation runs, to show the behaviour of our artificial economy along an
hypothetical scenario, or as averages across two-hundreds independent simulations to identify robust
emerging properties and to perform statistical testing across scenarios and policy experiments.
Before running policy experiments, the model has undergone an indirect calibration exercise. Then
we “empirically validate” the model, i.e. we study its capability to account for a large ensemble of macro
and micro stylized facts (see Fagiolo et al., 2019, and the Appendix for additional details). This is done
8
For more experiments on the short- and long-run impact of fiscal policies on public debt as well as economic dynamics,
see Dosi et al. (2015).
13Figure 2: Model behaviour under the “no innovation policy” baseline. Selected indicators are reported
for a single model run. GDP, Labour productivity, Real wage are in logs. Negative public deficit indicates
a surplus.
GDP Labor Productivity Unemployment rate
26
24 12 0.3
22 10
0.2
20
8
18 0.1
6
16
4 0.0
Real wage Public expenditures to GDP Deficit to GDP
0.20
0.1
10 0.15 0.0
0.10 −0.1
−0.2
5 0.05
−0.3
0.00
GDP, detrened (cyle) Consumption, detrended (cycle) Investments, detrended (cycle)
0.2
0.2
0.5
0.1 0.1
0.0
0.0 0.0
−0.5
−0.1
−0.1
−1.0
0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
periods
in the “no innovation policy” scenario, whose parameter configuration is reported in the Appendix.9
Figure 2 shows the dynamics of the model in the “no innovation policy” baseline relying on a sin-
gle model run, while Table 1 reports a series of summary statistics over the Monte Carlo ensemble.
The model robustly generates endogenous self-sustained growth patterns characterized by the pres-
ence of persistent fluctuations and rare crises. The positive trend in productivity and aggregate output
is driven by the innovation activity of capital-good firms and the processes of technological diffusion
occurring horizontally via the imitation activity of competitors, as well as vertically through the in-
vestment choices of consumption-good firms.
Simulation results also show the presence of fierce Schumpeterian competition taking place at the
microeconomic level. For instance, on average, slightly more than half of capital-good firms successfully
introduce an innovation or copy the technology of a competitor in every simulation step, while just
9
See also Lamperti (2018b,a), Guerini and Moneta (2017) and Lamperti et al. (2018).
14Table 1: Summary statistics for selected indicators in the “no innovation policy” baseline; 200 runs are
used. HHI stands for Hirschman-Herfindahl Index; Cap-Good indicates the Capital Good sector and
Cons-Good the Consumption good sector; Lik. stands for Likelihood; incr. and rad. for incremental
and radical respectively. Crises are defined as events where either GDP drops by more than 3% in a
single period or four consecutive periods of negative growth are observed.
Variable Mean St. Dev Variable Mean St. Dev
GDP growth 0.0268 0.0012 Unemployment 0.0610 0.0376
GDP volatility 0.0819 0.0005 Productivity growth 0.2581 0.0012
Deficit on GDP 0.0434 0.0553 HHI Cap-Good sector 0.6691 0.0601
Lik. of crises 0.0462 0.0399 HHI Cons-Good sector 0.0032 0.0001
Lik. of (incr.) innovation 0.571 0.0424 Lik. of imitation 0.6012 0.0502
Lik. of (incr.) inno. & imit. 0.218 0.0370 Lik. of (rad.) innov. 1.25 10−05 0.0000
one-fifth perform both activities. The likelihood of radical innovations is remarkably low, and only a
private firm is able to obtain one in a single run.
Government deficit averages around reasonable levels (4-6% of GDP) for a developed economy,
while displaying large spikes during crises, characterized by surges in unemployment. Beyond such
rare crises, whose likelihood is relatively low (around 5%), public finances often register a surplus,
guaranteeing the long run sustainability of debt.
The bottom panels in Figure 2 show the cyclical components of the GDP, consumption and invest-
ment time-series generated by our model. They show the presence of vibrant fluctuations in all series,
punctuated by deep downturns. Such fluctuations are genuinely endogenous, as no aggregate exoge-
nous shock is present in the model. In addition, consumption and investments are, respectively, more
and less volatile than output, in tune with empirical evidence (Stock and Watson, 1999; Napoletano
et al., 2006.
The K+S model is also able to account for a wide set of microeconomic empirical regularities con-
cerning e.g. firm size and growth-rate distributions, productivity dynamics, investment patterns. This
reflects the strong explanatory capabilities of agent-based models as discussed in Haldane and Turrell
(2019) and Dosi and Roventini (2019). Details about the empirically regularities replicated by the K+S
model are spelled out in the Appendix (see Table 7) and in Dosi et al. (2017).
Overall, our “no innovation policy” baseline reflects an economy where decentralized interactions
give rise to stable properties at the macroeconomic level (all standard deviations in Table 1 are rela-
tively low compared to the averages), with sustained growth and healthy public finances. Against such
background we test a series of policy regimes aimed at further stimulating innovation, productivity and
long-run growth, while maintaining public deficit and debt under control.
155 Innovation policy experiments
As emphasized in Section 2, innovation policy encompasses a variety of instruments, ranging from
monetary incentives such as R&D subsidies and tax credits (indirect interventions) to direct spending
in public research activities (for example, in the US, funding basic research through the National Sci-
ences Foundation as well as public organizations like DARPA of the US Department of Defense). In
this Section we rely on controlled simulation experiments to investigate the macroeconomic effects of
different policy instruments: Section 5.1 first describes the different policy interventions, while simu-
lation results are spelled out in Section 5.2. A sensitivity analysis of the main results can be found in
the Appendix (Table 9).
5.1 A “menu” of innovation policies
We consider five different types of innovations policies and we also experiment with ensembles of dif-
ferent interventions. Experiments I and II consider indirect policy interventions typical of the market
failure approach, whereas Experiments IV and V explore direct Government interventions and are akin
to the Entrepreneurial State framework. As an additional benchmark, we consider a scenario (Experi-
ment III) where public expenditures sustain only private consumption and hence cannot have a direct
influence on productivity growth.10
Experiment I: R&D subsidies. The Government provides a R&D subsidy to firms in order to increase their
research efforts. Larger R&D investments may increase the chances of discovering novel machines,
more efficient production techniques or, finally, they may speed up horizontal technological diffusion
via imitation of competitors. We assume that public subsidies qRD > 0 are proportional to firm’s past
spending in research and innovation (RDi ):
RDi (t) = υSi (t − 1) + qRD RDi (t − 1). (21)
Experiment II: investment tax discount. Under this intervention, consumption-good firms receive a
government-financed discount on their investments in novel capital goods, whose size - relative to
the price of the new machine - amounts to dT D . This policy is supposed to speed up technological
diffusion vertically, as consumption-good firms firms pay a lower prices whenever they replace cur-
rent machines with new ones embedding state-of-the-art technologies. Under this policy, the pay-back
period routine (cf. Eq. 14) becomes:
pnew (1 − dT D )
≤ b. (22)
ccon
j − cnew
Experiment III: public expenditures directed to private consumption. This experiment mimics a scenario
where public transfers boost household consumption expenditures. Of course, in this framework, they
10
In the model consumption positively affects demand expectations and thus expansionary investment. It may thus have,
via this channel, a positive effect on R&D in the capital good sector, which depends on past sales. Nevertheless, its impact is
expected to be lower compared to R&D subsidies and direct government innovation policies.
16Figure 3: Experiment IV: knowledge diffusion by the public firm
B public
firm
private
firm
A
Knowledge diffusion
do not directly affect the innovation and investment decisions of firms, but they might increase pro-
ductivity growth via more sustained levels of aggregate demand.
Experiment IV: a public capital-good firm. In an Entrepreneurial State framework, new public entities
are created to shape the innovation landscape by engaging and coordinating research in given fields
and diffusing the relevant knowledge to facilitate technological progress (see sections 1 and 2). In
this experiment, the government creates and fund a public firm in the capital-good sector. Similarly to
privately owned firms, the new public firm satisfies the demand of machines coming from consumption-
good firms and performs innovation and imitation activities. However, four key differences apply: i) the
public firm allocates all its profits (Πpf ) to R&D; ii) it is bailed out by the government in case of failure
(negative net liquid assets); iii) it can receive additional funds from the government (IP ) to perform
extra research activities; and iv) it fosters the diffusion of its technology to its competitors which can
freely imitate it if their cumulated knowledge is sufficiently large. In particular, the R&D expenditure
of the public firm (pf ) amounts to:
RDpf (t) = υSpf (t − 1) + Πpf (t − 1) + IP (t). (23)
Any capital-good firms i can freely imitate the public firm if its (normalized) technological distance -
which stems from the history and direction of its innovations - from the public firm (N T Dpf,j , cf. Eq.
9) is smaller than a fixed threshold φ ∈ (0, 1). Figure 3 shows a stylized representation of such a “local”
process of knowledge diffusion. Obviously, private firms will decide whether to adopt the technology
of the public firm only if it is convenient on the basis of the routine expressed by Equation (10).
Experiment V: a national research laboratory. The last experiment captures another essential feature of
an Entrepreneurial State, i.e. the creation and funding of public institutions that discover radical in-
novations enlarging technological opportunities in the economy (as for national research laboratories
and the Internet, see section 2), while bearing the risks and the costs of such ventures. In particular,
17Table 2: Results from Experiment I (R&D subsidies). Rows reports the average relative performance of
each experiment with respect to the “no innovation policy” baseline (Baseline) over 200 Monte Carlo
runs; for example 1.2 indicates that the experiment has produced an average value of the relevant
statistic that is 20% higher than in the baseline. Symbol * indicates a statistical significant difference
between the experiment and the baseline at 5% as resulting from a t-test on the means. GDP vol. stands
for GDP volatility as proxied by the standard deviation of the growth process; Unempl. stands for
unemployment and empl. for employment; Deficit and Fiscal cost are expressed as relative to GDP.
GDP growth GDP vol. Unempl. Periods full empl. Deficit Fiscal cost
Baseline 2.68% 0.08 6.10% 16% 4.34% 0.00
Size of the subsidy
5% 1.04 1.01 0.98 1.04 1.25 0.9% *
10% 1.08 * 1.02 0.98 1.08 1.39 * 2.2% *
15% 1.10 * 0.97 0.96 1.17 * 1.14 * 2.6% *
30% 1.18 * 0.99 0.95 1.37 * 0.94 6.4% *
we introduce a national research lab (NRL) that (i) performs basic research but does not produce; (ii)
takes stock of all the knowledge developed in the economy, (iii) tries to enlarge the set of technological
opportunities available for capital-good firms through the discovery of radical innovations (see Section
3.1). At each time step, the NRL receives public funding form the government to perform its research
activities. Further, as it is a purely research-oriented organization, it is able to exploit the entire body
of knowledge available in the economy to perform its research. Hence, the discovery of a radical in-
novation by the NRL is assumed to depend on its cumulative search efforts (CRDpublic ), as well as on
those performed by capital-good firms (CRDi ):11
P
i CRDi (t) + CRDpublic (t) 1
PNRIRL (t) =f x|x = = . (24)
GDP (t) 1+ eη1 (x−η2 )
Differently from private firms (see section 3.1), a NRL that discovers a radical innovation, also provides
free access to the new technological opportunities it involves, de facto moving the distribution of in-
novative possibilities for the whole economy.12
5.2 Simulation results
To ensure the comparability of results across the different policy experiments, we keep constant the
fiscal cost of the innovation policies in the various regimes. In particular, we first perform Experiment
I (R&D subsidy) by setting the size of the subsidy (qRD ∈ {5%, 10%, 15%, 30%}). Then, we inspect the
11
To the contrary, the probability that private firms discover a radical innovation depends on own cumulative R&D and
the R&D expenditures by the NRL, if any. See Equation 12.
12
In the current set-up, we cannot study mission-oriented innovation policies directed to specific missions, as the model
does not allow for multiple industries. Hence, we cannot study how such policies trigger the direction of technical change
through the emergence of new sectors and markets. We leave such developments to future research (see also our discussion
in Section 6).
18You can also read
